A New Sampling Algorithm in Particle Filter for Geotechnical Analysis

Abstract

This paper discusses the applicability of the particle filter (PF) algorithms to geotechnical analysis through some numerical tests. Although several types of the PF algorithms have been proposed so far, this study focuses on three typical PF algorithms: sequential importance resampling (SIR), sequential importance sampling (SIS), and merging particle filter (MPF). First, a geotechnical parameter is identified using the three algorithms in both total stress and soil-water coupled analyses, and the effectiveness of each algorithm is investigated. The test results clarify that (1)SIS can be applied to non-Markov dynamics such as elasto-plastic problems, but degeneration problems are often encountered, and (2)MPF can avoid the degeneration problems, but it cannot be applied to non-Markov dynamics. To overcome the dilemma, an algorithm which can treat non-Markov dynamics and solve the degeneration problems is newly proposed. The proposed algorithm is applied to an element test, and the performance is demonstrated experimentally.